The traditional "spinning platter" hard disk drive tested was capable of a maximum sustained transfer rate of approximately 4GB/minute. The hard drive that was tested was much slower than a 1.5Gbs SATA connection, meaning that the bottleneck was the disk drive.
The chart below shows how long it took to copy 10GB of data from a hard disk drive to /dev/null with varying buffer sizes. The results show that buffer sizes less than 4096 bytes were inefficient, with a buffer size of 8 bytes as very inefficient. This indicates that small buffer sizes result in a poor ratio of data to general protocol traffic on the SATA bus. However, buffer sizes greater than 2048 bytes did not appear to have a significant impact for this disk drive.
This chart displays the impact of differing buffer sizes on acquisition time of a 750 GB disk drive via SATA. (Note that the X-axis is linear.) Buffer sizes greater than 4096 bytes are clustered fairly close together. Buffer sizes that are less than 512 bytes were significantly less efficient than 4096 bytes.
The percentage over baseline chart treats the fastest copying time as the goal. For example, testing of the 750 GB hard disk drive revealed that a buffer size of 4096 bytes was the most efficient. The time necessary to copy 10GB using a buffer size of 4096 bytes was 112 seconds. Thus, this time became the baseline. The time necessary to copy 10GB with different buffer sizes are then compared to the 112 seconds.